1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Generic socket support routines. Memory allocators, socket lock/release 7 * handler for protocols to use and generic option handler. 8 * 9 * 10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $ 11 * 12 * Authors: Ross Biro 13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 14 * Florian La Roche, <flla@stud.uni-sb.de> 15 * Alan Cox, <A.Cox@swansea.ac.uk> 16 * 17 * Fixes: 18 * Alan Cox : Numerous verify_area() problems 19 * Alan Cox : Connecting on a connecting socket 20 * now returns an error for tcp. 21 * Alan Cox : sock->protocol is set correctly. 22 * and is not sometimes left as 0. 23 * Alan Cox : connect handles icmp errors on a 24 * connect properly. Unfortunately there 25 * is a restart syscall nasty there. I 26 * can't match BSD without hacking the C 27 * library. Ideas urgently sought! 28 * Alan Cox : Disallow bind() to addresses that are 29 * not ours - especially broadcast ones!! 30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 32 * instead they leave that for the DESTROY timer. 33 * Alan Cox : Clean up error flag in accept 34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 35 * was buggy. Put a remove_sock() in the handler 36 * for memory when we hit 0. Also altered the timer 37 * code. The ACK stuff can wait and needs major 38 * TCP layer surgery. 39 * Alan Cox : Fixed TCP ack bug, removed remove sock 40 * and fixed timer/inet_bh race. 41 * Alan Cox : Added zapped flag for TCP 42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 47 * Rick Sladkey : Relaxed UDP rules for matching packets. 48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 49 * Pauline Middelink : identd support 50 * Alan Cox : Fixed connect() taking signals I think. 51 * Alan Cox : SO_LINGER supported 52 * Alan Cox : Error reporting fixes 53 * Anonymous : inet_create tidied up (sk->reuse setting) 54 * Alan Cox : inet sockets don't set sk->type! 55 * Alan Cox : Split socket option code 56 * Alan Cox : Callbacks 57 * Alan Cox : Nagle flag for Charles & Johannes stuff 58 * Alex : Removed restriction on inet fioctl 59 * Alan Cox : Splitting INET from NET core 60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 62 * Alan Cox : Split IP from generic code 63 * Alan Cox : New kfree_skbmem() 64 * Alan Cox : Make SO_DEBUG superuser only. 65 * Alan Cox : Allow anyone to clear SO_DEBUG 66 * (compatibility fix) 67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 68 * Alan Cox : Allocator for a socket is settable. 69 * Alan Cox : SO_ERROR includes soft errors. 70 * Alan Cox : Allow NULL arguments on some SO_ opts 71 * Alan Cox : Generic socket allocation to make hooks 72 * easier (suggested by Craig Metz). 73 * Michael Pall : SO_ERROR returns positive errno again 74 * Steve Whitehouse: Added default destructor to free 75 * protocol private data. 76 * Steve Whitehouse: Added various other default routines 77 * common to several socket families. 78 * Chris Evans : Call suser() check last on F_SETOWN 79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 81 * Andi Kleen : Fix write_space callback 82 * Chris Evans : Security fixes - signedness again 83 * Arnaldo C. Melo : cleanups, use skb_queue_purge 84 * 85 * To Fix: 86 * 87 * 88 * This program is free software; you can redistribute it and/or 89 * modify it under the terms of the GNU General Public License 90 * as published by the Free Software Foundation; either version 91 * 2 of the License, or (at your option) any later version. 92 */ 93 94 #include <linux/capability.h> 95 #include <linux/errno.h> 96 #include <linux/types.h> 97 #include <linux/socket.h> 98 #include <linux/in.h> 99 #include <linux/kernel.h> 100 #include <linux/module.h> 101 #include <linux/proc_fs.h> 102 #include <linux/seq_file.h> 103 #include <linux/sched.h> 104 #include <linux/timer.h> 105 #include <linux/string.h> 106 #include <linux/sockios.h> 107 #include <linux/net.h> 108 #include <linux/mm.h> 109 #include <linux/slab.h> 110 #include <linux/interrupt.h> 111 #include <linux/poll.h> 112 #include <linux/tcp.h> 113 #include <linux/init.h> 114 #include <linux/highmem.h> 115 116 #include <asm/uaccess.h> 117 #include <asm/system.h> 118 119 #include <linux/netdevice.h> 120 #include <net/protocol.h> 121 #include <linux/skbuff.h> 122 #include <net/request_sock.h> 123 #include <net/sock.h> 124 #include <net/xfrm.h> 125 #include <linux/ipsec.h> 126 127 #include <linux/filter.h> 128 129 #ifdef CONFIG_INET 130 #include <net/tcp.h> 131 #endif 132 133 /* 134 * Each address family might have different locking rules, so we have 135 * one slock key per address family: 136 */ 137 static struct lock_class_key af_family_keys[AF_MAX]; 138 static struct lock_class_key af_family_slock_keys[AF_MAX]; 139 140 #ifdef CONFIG_DEBUG_LOCK_ALLOC 141 /* 142 * Make lock validator output more readable. (we pre-construct these 143 * strings build-time, so that runtime initialization of socket 144 * locks is fast): 145 */ 146 static const char *af_family_key_strings[AF_MAX+1] = { 147 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 148 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 149 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 150 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 151 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 152 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 153 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 154 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 155 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 156 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" , 157 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX" 158 }; 159 static const char *af_family_slock_key_strings[AF_MAX+1] = { 160 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 161 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 162 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 163 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 164 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 165 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 166 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 167 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" , 168 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 169 "slock-27" , "slock-28" , "slock-29" , 170 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX" 171 }; 172 #endif 173 174 /* 175 * sk_callback_lock locking rules are per-address-family, 176 * so split the lock classes by using a per-AF key: 177 */ 178 static struct lock_class_key af_callback_keys[AF_MAX]; 179 180 /* Take into consideration the size of the struct sk_buff overhead in the 181 * determination of these values, since that is non-constant across 182 * platforms. This makes socket queueing behavior and performance 183 * not depend upon such differences. 184 */ 185 #define _SK_MEM_PACKETS 256 186 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256) 187 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 188 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 189 190 /* Run time adjustable parameters. */ 191 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 192 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 193 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 194 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 195 196 /* Maximal space eaten by iovec or ancilliary data plus some space */ 197 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 198 199 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 200 { 201 struct timeval tv; 202 203 if (optlen < sizeof(tv)) 204 return -EINVAL; 205 if (copy_from_user(&tv, optval, sizeof(tv))) 206 return -EFAULT; 207 208 *timeo_p = MAX_SCHEDULE_TIMEOUT; 209 if (tv.tv_sec == 0 && tv.tv_usec == 0) 210 return 0; 211 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 212 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 213 return 0; 214 } 215 216 static void sock_warn_obsolete_bsdism(const char *name) 217 { 218 static int warned; 219 static char warncomm[TASK_COMM_LEN]; 220 if (strcmp(warncomm, current->comm) && warned < 5) { 221 strcpy(warncomm, current->comm); 222 printk(KERN_WARNING "process `%s' is using obsolete " 223 "%s SO_BSDCOMPAT\n", warncomm, name); 224 warned++; 225 } 226 } 227 228 static void sock_disable_timestamp(struct sock *sk) 229 { 230 if (sock_flag(sk, SOCK_TIMESTAMP)) { 231 sock_reset_flag(sk, SOCK_TIMESTAMP); 232 net_disable_timestamp(); 233 } 234 } 235 236 237 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 238 { 239 int err = 0; 240 int skb_len; 241 242 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces 243 number of warnings when compiling with -W --ANK 244 */ 245 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 246 (unsigned)sk->sk_rcvbuf) { 247 err = -ENOMEM; 248 goto out; 249 } 250 251 err = sk_filter(sk, skb); 252 if (err) 253 goto out; 254 255 skb->dev = NULL; 256 skb_set_owner_r(skb, sk); 257 258 /* Cache the SKB length before we tack it onto the receive 259 * queue. Once it is added it no longer belongs to us and 260 * may be freed by other threads of control pulling packets 261 * from the queue. 262 */ 263 skb_len = skb->len; 264 265 skb_queue_tail(&sk->sk_receive_queue, skb); 266 267 if (!sock_flag(sk, SOCK_DEAD)) 268 sk->sk_data_ready(sk, skb_len); 269 out: 270 return err; 271 } 272 EXPORT_SYMBOL(sock_queue_rcv_skb); 273 274 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 275 { 276 int rc = NET_RX_SUCCESS; 277 278 if (sk_filter(sk, skb)) 279 goto discard_and_relse; 280 281 skb->dev = NULL; 282 283 if (nested) 284 bh_lock_sock_nested(sk); 285 else 286 bh_lock_sock(sk); 287 if (!sock_owned_by_user(sk)) { 288 /* 289 * trylock + unlock semantics: 290 */ 291 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 292 293 rc = sk->sk_backlog_rcv(sk, skb); 294 295 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 296 } else 297 sk_add_backlog(sk, skb); 298 bh_unlock_sock(sk); 299 out: 300 sock_put(sk); 301 return rc; 302 discard_and_relse: 303 kfree_skb(skb); 304 goto out; 305 } 306 EXPORT_SYMBOL(sk_receive_skb); 307 308 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 309 { 310 struct dst_entry *dst = sk->sk_dst_cache; 311 312 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 313 sk->sk_dst_cache = NULL; 314 dst_release(dst); 315 return NULL; 316 } 317 318 return dst; 319 } 320 EXPORT_SYMBOL(__sk_dst_check); 321 322 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 323 { 324 struct dst_entry *dst = sk_dst_get(sk); 325 326 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 327 sk_dst_reset(sk); 328 dst_release(dst); 329 return NULL; 330 } 331 332 return dst; 333 } 334 EXPORT_SYMBOL(sk_dst_check); 335 336 /* 337 * This is meant for all protocols to use and covers goings on 338 * at the socket level. Everything here is generic. 339 */ 340 341 int sock_setsockopt(struct socket *sock, int level, int optname, 342 char __user *optval, int optlen) 343 { 344 struct sock *sk=sock->sk; 345 struct sk_filter *filter; 346 int val; 347 int valbool; 348 struct linger ling; 349 int ret = 0; 350 351 /* 352 * Options without arguments 353 */ 354 355 #ifdef SO_DONTLINGER /* Compatibility item... */ 356 if (optname == SO_DONTLINGER) { 357 lock_sock(sk); 358 sock_reset_flag(sk, SOCK_LINGER); 359 release_sock(sk); 360 return 0; 361 } 362 #endif 363 364 if(optlen<sizeof(int)) 365 return(-EINVAL); 366 367 if (get_user(val, (int __user *)optval)) 368 return -EFAULT; 369 370 valbool = val?1:0; 371 372 lock_sock(sk); 373 374 switch(optname) 375 { 376 case SO_DEBUG: 377 if(val && !capable(CAP_NET_ADMIN)) 378 { 379 ret = -EACCES; 380 } 381 else if (valbool) 382 sock_set_flag(sk, SOCK_DBG); 383 else 384 sock_reset_flag(sk, SOCK_DBG); 385 break; 386 case SO_REUSEADDR: 387 sk->sk_reuse = valbool; 388 break; 389 case SO_TYPE: 390 case SO_ERROR: 391 ret = -ENOPROTOOPT; 392 break; 393 case SO_DONTROUTE: 394 if (valbool) 395 sock_set_flag(sk, SOCK_LOCALROUTE); 396 else 397 sock_reset_flag(sk, SOCK_LOCALROUTE); 398 break; 399 case SO_BROADCAST: 400 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 401 break; 402 case SO_SNDBUF: 403 /* Don't error on this BSD doesn't and if you think 404 about it this is right. Otherwise apps have to 405 play 'guess the biggest size' games. RCVBUF/SNDBUF 406 are treated in BSD as hints */ 407 408 if (val > sysctl_wmem_max) 409 val = sysctl_wmem_max; 410 set_sndbuf: 411 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 412 if ((val * 2) < SOCK_MIN_SNDBUF) 413 sk->sk_sndbuf = SOCK_MIN_SNDBUF; 414 else 415 sk->sk_sndbuf = val * 2; 416 417 /* 418 * Wake up sending tasks if we 419 * upped the value. 420 */ 421 sk->sk_write_space(sk); 422 break; 423 424 case SO_SNDBUFFORCE: 425 if (!capable(CAP_NET_ADMIN)) { 426 ret = -EPERM; 427 break; 428 } 429 goto set_sndbuf; 430 431 case SO_RCVBUF: 432 /* Don't error on this BSD doesn't and if you think 433 about it this is right. Otherwise apps have to 434 play 'guess the biggest size' games. RCVBUF/SNDBUF 435 are treated in BSD as hints */ 436 437 if (val > sysctl_rmem_max) 438 val = sysctl_rmem_max; 439 set_rcvbuf: 440 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 441 /* 442 * We double it on the way in to account for 443 * "struct sk_buff" etc. overhead. Applications 444 * assume that the SO_RCVBUF setting they make will 445 * allow that much actual data to be received on that 446 * socket. 447 * 448 * Applications are unaware that "struct sk_buff" and 449 * other overheads allocate from the receive buffer 450 * during socket buffer allocation. 451 * 452 * And after considering the possible alternatives, 453 * returning the value we actually used in getsockopt 454 * is the most desirable behavior. 455 */ 456 if ((val * 2) < SOCK_MIN_RCVBUF) 457 sk->sk_rcvbuf = SOCK_MIN_RCVBUF; 458 else 459 sk->sk_rcvbuf = val * 2; 460 break; 461 462 case SO_RCVBUFFORCE: 463 if (!capable(CAP_NET_ADMIN)) { 464 ret = -EPERM; 465 break; 466 } 467 goto set_rcvbuf; 468 469 case SO_KEEPALIVE: 470 #ifdef CONFIG_INET 471 if (sk->sk_protocol == IPPROTO_TCP) 472 tcp_set_keepalive(sk, valbool); 473 #endif 474 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 475 break; 476 477 case SO_OOBINLINE: 478 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 479 break; 480 481 case SO_NO_CHECK: 482 sk->sk_no_check = valbool; 483 break; 484 485 case SO_PRIORITY: 486 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 487 sk->sk_priority = val; 488 else 489 ret = -EPERM; 490 break; 491 492 case SO_LINGER: 493 if(optlen<sizeof(ling)) { 494 ret = -EINVAL; /* 1003.1g */ 495 break; 496 } 497 if (copy_from_user(&ling,optval,sizeof(ling))) { 498 ret = -EFAULT; 499 break; 500 } 501 if (!ling.l_onoff) 502 sock_reset_flag(sk, SOCK_LINGER); 503 else { 504 #if (BITS_PER_LONG == 32) 505 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 506 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 507 else 508 #endif 509 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 510 sock_set_flag(sk, SOCK_LINGER); 511 } 512 break; 513 514 case SO_BSDCOMPAT: 515 sock_warn_obsolete_bsdism("setsockopt"); 516 break; 517 518 case SO_PASSCRED: 519 if (valbool) 520 set_bit(SOCK_PASSCRED, &sock->flags); 521 else 522 clear_bit(SOCK_PASSCRED, &sock->flags); 523 break; 524 525 case SO_TIMESTAMP: 526 if (valbool) { 527 sock_set_flag(sk, SOCK_RCVTSTAMP); 528 sock_enable_timestamp(sk); 529 } else 530 sock_reset_flag(sk, SOCK_RCVTSTAMP); 531 break; 532 533 case SO_RCVLOWAT: 534 if (val < 0) 535 val = INT_MAX; 536 sk->sk_rcvlowat = val ? : 1; 537 break; 538 539 case SO_RCVTIMEO: 540 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 541 break; 542 543 case SO_SNDTIMEO: 544 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 545 break; 546 547 #ifdef CONFIG_NETDEVICES 548 case SO_BINDTODEVICE: 549 { 550 char devname[IFNAMSIZ]; 551 552 /* Sorry... */ 553 if (!capable(CAP_NET_RAW)) { 554 ret = -EPERM; 555 break; 556 } 557 558 /* Bind this socket to a particular device like "eth0", 559 * as specified in the passed interface name. If the 560 * name is "" or the option length is zero the socket 561 * is not bound. 562 */ 563 564 if (!valbool) { 565 sk->sk_bound_dev_if = 0; 566 } else { 567 if (optlen > IFNAMSIZ - 1) 568 optlen = IFNAMSIZ - 1; 569 memset(devname, 0, sizeof(devname)); 570 if (copy_from_user(devname, optval, optlen)) { 571 ret = -EFAULT; 572 break; 573 } 574 575 /* Remove any cached route for this socket. */ 576 sk_dst_reset(sk); 577 578 if (devname[0] == '\0') { 579 sk->sk_bound_dev_if = 0; 580 } else { 581 struct net_device *dev = dev_get_by_name(devname); 582 if (!dev) { 583 ret = -ENODEV; 584 break; 585 } 586 sk->sk_bound_dev_if = dev->ifindex; 587 dev_put(dev); 588 } 589 } 590 break; 591 } 592 #endif 593 594 595 case SO_ATTACH_FILTER: 596 ret = -EINVAL; 597 if (optlen == sizeof(struct sock_fprog)) { 598 struct sock_fprog fprog; 599 600 ret = -EFAULT; 601 if (copy_from_user(&fprog, optval, sizeof(fprog))) 602 break; 603 604 ret = sk_attach_filter(&fprog, sk); 605 } 606 break; 607 608 case SO_DETACH_FILTER: 609 rcu_read_lock_bh(); 610 filter = rcu_dereference(sk->sk_filter); 611 if (filter) { 612 rcu_assign_pointer(sk->sk_filter, NULL); 613 sk_filter_release(sk, filter); 614 rcu_read_unlock_bh(); 615 break; 616 } 617 rcu_read_unlock_bh(); 618 ret = -ENONET; 619 break; 620 621 case SO_PASSSEC: 622 if (valbool) 623 set_bit(SOCK_PASSSEC, &sock->flags); 624 else 625 clear_bit(SOCK_PASSSEC, &sock->flags); 626 break; 627 628 /* We implement the SO_SNDLOWAT etc to 629 not be settable (1003.1g 5.3) */ 630 default: 631 ret = -ENOPROTOOPT; 632 break; 633 } 634 release_sock(sk); 635 return ret; 636 } 637 638 639 int sock_getsockopt(struct socket *sock, int level, int optname, 640 char __user *optval, int __user *optlen) 641 { 642 struct sock *sk = sock->sk; 643 644 union 645 { 646 int val; 647 struct linger ling; 648 struct timeval tm; 649 } v; 650 651 unsigned int lv = sizeof(int); 652 int len; 653 654 if(get_user(len,optlen)) 655 return -EFAULT; 656 if(len < 0) 657 return -EINVAL; 658 659 switch(optname) 660 { 661 case SO_DEBUG: 662 v.val = sock_flag(sk, SOCK_DBG); 663 break; 664 665 case SO_DONTROUTE: 666 v.val = sock_flag(sk, SOCK_LOCALROUTE); 667 break; 668 669 case SO_BROADCAST: 670 v.val = !!sock_flag(sk, SOCK_BROADCAST); 671 break; 672 673 case SO_SNDBUF: 674 v.val = sk->sk_sndbuf; 675 break; 676 677 case SO_RCVBUF: 678 v.val = sk->sk_rcvbuf; 679 break; 680 681 case SO_REUSEADDR: 682 v.val = sk->sk_reuse; 683 break; 684 685 case SO_KEEPALIVE: 686 v.val = !!sock_flag(sk, SOCK_KEEPOPEN); 687 break; 688 689 case SO_TYPE: 690 v.val = sk->sk_type; 691 break; 692 693 case SO_ERROR: 694 v.val = -sock_error(sk); 695 if(v.val==0) 696 v.val = xchg(&sk->sk_err_soft, 0); 697 break; 698 699 case SO_OOBINLINE: 700 v.val = !!sock_flag(sk, SOCK_URGINLINE); 701 break; 702 703 case SO_NO_CHECK: 704 v.val = sk->sk_no_check; 705 break; 706 707 case SO_PRIORITY: 708 v.val = sk->sk_priority; 709 break; 710 711 case SO_LINGER: 712 lv = sizeof(v.ling); 713 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER); 714 v.ling.l_linger = sk->sk_lingertime / HZ; 715 break; 716 717 case SO_BSDCOMPAT: 718 sock_warn_obsolete_bsdism("getsockopt"); 719 break; 720 721 case SO_TIMESTAMP: 722 v.val = sock_flag(sk, SOCK_RCVTSTAMP); 723 break; 724 725 case SO_RCVTIMEO: 726 lv=sizeof(struct timeval); 727 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 728 v.tm.tv_sec = 0; 729 v.tm.tv_usec = 0; 730 } else { 731 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 732 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 733 } 734 break; 735 736 case SO_SNDTIMEO: 737 lv=sizeof(struct timeval); 738 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 739 v.tm.tv_sec = 0; 740 v.tm.tv_usec = 0; 741 } else { 742 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 743 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 744 } 745 break; 746 747 case SO_RCVLOWAT: 748 v.val = sk->sk_rcvlowat; 749 break; 750 751 case SO_SNDLOWAT: 752 v.val=1; 753 break; 754 755 case SO_PASSCRED: 756 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0; 757 break; 758 759 case SO_PEERCRED: 760 if (len > sizeof(sk->sk_peercred)) 761 len = sizeof(sk->sk_peercred); 762 if (copy_to_user(optval, &sk->sk_peercred, len)) 763 return -EFAULT; 764 goto lenout; 765 766 case SO_PEERNAME: 767 { 768 char address[128]; 769 770 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 771 return -ENOTCONN; 772 if (lv < len) 773 return -EINVAL; 774 if (copy_to_user(optval, address, len)) 775 return -EFAULT; 776 goto lenout; 777 } 778 779 /* Dubious BSD thing... Probably nobody even uses it, but 780 * the UNIX standard wants it for whatever reason... -DaveM 781 */ 782 case SO_ACCEPTCONN: 783 v.val = sk->sk_state == TCP_LISTEN; 784 break; 785 786 case SO_PASSSEC: 787 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0; 788 break; 789 790 case SO_PEERSEC: 791 return security_socket_getpeersec_stream(sock, optval, optlen, len); 792 793 default: 794 return(-ENOPROTOOPT); 795 } 796 if (len > lv) 797 len = lv; 798 if (copy_to_user(optval, &v, len)) 799 return -EFAULT; 800 lenout: 801 if (put_user(len, optlen)) 802 return -EFAULT; 803 return 0; 804 } 805 806 /* 807 * Initialize an sk_lock. 808 * 809 * (We also register the sk_lock with the lock validator.) 810 */ 811 static void inline sock_lock_init(struct sock *sk) 812 { 813 sock_lock_init_class_and_name(sk, 814 af_family_slock_key_strings[sk->sk_family], 815 af_family_slock_keys + sk->sk_family, 816 af_family_key_strings[sk->sk_family], 817 af_family_keys + sk->sk_family); 818 } 819 820 /** 821 * sk_alloc - All socket objects are allocated here 822 * @family: protocol family 823 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 824 * @prot: struct proto associated with this new sock instance 825 * @zero_it: if we should zero the newly allocated sock 826 */ 827 struct sock *sk_alloc(int family, gfp_t priority, 828 struct proto *prot, int zero_it) 829 { 830 struct sock *sk = NULL; 831 struct kmem_cache *slab = prot->slab; 832 833 if (slab != NULL) 834 sk = kmem_cache_alloc(slab, priority); 835 else 836 sk = kmalloc(prot->obj_size, priority); 837 838 if (sk) { 839 if (zero_it) { 840 memset(sk, 0, prot->obj_size); 841 sk->sk_family = family; 842 /* 843 * See comment in struct sock definition to understand 844 * why we need sk_prot_creator -acme 845 */ 846 sk->sk_prot = sk->sk_prot_creator = prot; 847 sock_lock_init(sk); 848 } 849 850 if (security_sk_alloc(sk, family, priority)) 851 goto out_free; 852 853 if (!try_module_get(prot->owner)) 854 goto out_free; 855 } 856 return sk; 857 858 out_free: 859 if (slab != NULL) 860 kmem_cache_free(slab, sk); 861 else 862 kfree(sk); 863 return NULL; 864 } 865 866 void sk_free(struct sock *sk) 867 { 868 struct sk_filter *filter; 869 struct module *owner = sk->sk_prot_creator->owner; 870 871 if (sk->sk_destruct) 872 sk->sk_destruct(sk); 873 874 filter = rcu_dereference(sk->sk_filter); 875 if (filter) { 876 sk_filter_release(sk, filter); 877 rcu_assign_pointer(sk->sk_filter, NULL); 878 } 879 880 sock_disable_timestamp(sk); 881 882 if (atomic_read(&sk->sk_omem_alloc)) 883 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n", 884 __FUNCTION__, atomic_read(&sk->sk_omem_alloc)); 885 886 security_sk_free(sk); 887 if (sk->sk_prot_creator->slab != NULL) 888 kmem_cache_free(sk->sk_prot_creator->slab, sk); 889 else 890 kfree(sk); 891 module_put(owner); 892 } 893 894 struct sock *sk_clone(const struct sock *sk, const gfp_t priority) 895 { 896 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0); 897 898 if (newsk != NULL) { 899 struct sk_filter *filter; 900 901 sock_copy(newsk, sk); 902 903 /* SANITY */ 904 sk_node_init(&newsk->sk_node); 905 sock_lock_init(newsk); 906 bh_lock_sock(newsk); 907 908 atomic_set(&newsk->sk_rmem_alloc, 0); 909 atomic_set(&newsk->sk_wmem_alloc, 0); 910 atomic_set(&newsk->sk_omem_alloc, 0); 911 skb_queue_head_init(&newsk->sk_receive_queue); 912 skb_queue_head_init(&newsk->sk_write_queue); 913 #ifdef CONFIG_NET_DMA 914 skb_queue_head_init(&newsk->sk_async_wait_queue); 915 #endif 916 917 rwlock_init(&newsk->sk_dst_lock); 918 rwlock_init(&newsk->sk_callback_lock); 919 lockdep_set_class(&newsk->sk_callback_lock, 920 af_callback_keys + newsk->sk_family); 921 922 newsk->sk_dst_cache = NULL; 923 newsk->sk_wmem_queued = 0; 924 newsk->sk_forward_alloc = 0; 925 newsk->sk_send_head = NULL; 926 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 927 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 928 929 sock_reset_flag(newsk, SOCK_DONE); 930 skb_queue_head_init(&newsk->sk_error_queue); 931 932 filter = newsk->sk_filter; 933 if (filter != NULL) 934 sk_filter_charge(newsk, filter); 935 936 if (unlikely(xfrm_sk_clone_policy(newsk))) { 937 /* It is still raw copy of parent, so invalidate 938 * destructor and make plain sk_free() */ 939 newsk->sk_destruct = NULL; 940 sk_free(newsk); 941 newsk = NULL; 942 goto out; 943 } 944 945 newsk->sk_err = 0; 946 newsk->sk_priority = 0; 947 atomic_set(&newsk->sk_refcnt, 2); 948 949 /* 950 * Increment the counter in the same struct proto as the master 951 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 952 * is the same as sk->sk_prot->socks, as this field was copied 953 * with memcpy). 954 * 955 * This _changes_ the previous behaviour, where 956 * tcp_create_openreq_child always was incrementing the 957 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 958 * to be taken into account in all callers. -acme 959 */ 960 sk_refcnt_debug_inc(newsk); 961 newsk->sk_socket = NULL; 962 newsk->sk_sleep = NULL; 963 964 if (newsk->sk_prot->sockets_allocated) 965 atomic_inc(newsk->sk_prot->sockets_allocated); 966 } 967 out: 968 return newsk; 969 } 970 971 EXPORT_SYMBOL_GPL(sk_clone); 972 973 void __init sk_init(void) 974 { 975 if (num_physpages <= 4096) { 976 sysctl_wmem_max = 32767; 977 sysctl_rmem_max = 32767; 978 sysctl_wmem_default = 32767; 979 sysctl_rmem_default = 32767; 980 } else if (num_physpages >= 131072) { 981 sysctl_wmem_max = 131071; 982 sysctl_rmem_max = 131071; 983 } 984 } 985 986 /* 987 * Simple resource managers for sockets. 988 */ 989 990 991 /* 992 * Write buffer destructor automatically called from kfree_skb. 993 */ 994 void sock_wfree(struct sk_buff *skb) 995 { 996 struct sock *sk = skb->sk; 997 998 /* In case it might be waiting for more memory. */ 999 atomic_sub(skb->truesize, &sk->sk_wmem_alloc); 1000 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) 1001 sk->sk_write_space(sk); 1002 sock_put(sk); 1003 } 1004 1005 /* 1006 * Read buffer destructor automatically called from kfree_skb. 1007 */ 1008 void sock_rfree(struct sk_buff *skb) 1009 { 1010 struct sock *sk = skb->sk; 1011 1012 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1013 } 1014 1015 1016 int sock_i_uid(struct sock *sk) 1017 { 1018 int uid; 1019 1020 read_lock(&sk->sk_callback_lock); 1021 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1022 read_unlock(&sk->sk_callback_lock); 1023 return uid; 1024 } 1025 1026 unsigned long sock_i_ino(struct sock *sk) 1027 { 1028 unsigned long ino; 1029 1030 read_lock(&sk->sk_callback_lock); 1031 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1032 read_unlock(&sk->sk_callback_lock); 1033 return ino; 1034 } 1035 1036 /* 1037 * Allocate a skb from the socket's send buffer. 1038 */ 1039 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1040 gfp_t priority) 1041 { 1042 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1043 struct sk_buff * skb = alloc_skb(size, priority); 1044 if (skb) { 1045 skb_set_owner_w(skb, sk); 1046 return skb; 1047 } 1048 } 1049 return NULL; 1050 } 1051 1052 /* 1053 * Allocate a skb from the socket's receive buffer. 1054 */ 1055 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1056 gfp_t priority) 1057 { 1058 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1059 struct sk_buff *skb = alloc_skb(size, priority); 1060 if (skb) { 1061 skb_set_owner_r(skb, sk); 1062 return skb; 1063 } 1064 } 1065 return NULL; 1066 } 1067 1068 /* 1069 * Allocate a memory block from the socket's option memory buffer. 1070 */ 1071 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1072 { 1073 if ((unsigned)size <= sysctl_optmem_max && 1074 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1075 void *mem; 1076 /* First do the add, to avoid the race if kmalloc 1077 * might sleep. 1078 */ 1079 atomic_add(size, &sk->sk_omem_alloc); 1080 mem = kmalloc(size, priority); 1081 if (mem) 1082 return mem; 1083 atomic_sub(size, &sk->sk_omem_alloc); 1084 } 1085 return NULL; 1086 } 1087 1088 /* 1089 * Free an option memory block. 1090 */ 1091 void sock_kfree_s(struct sock *sk, void *mem, int size) 1092 { 1093 kfree(mem); 1094 atomic_sub(size, &sk->sk_omem_alloc); 1095 } 1096 1097 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1098 I think, these locks should be removed for datagram sockets. 1099 */ 1100 static long sock_wait_for_wmem(struct sock * sk, long timeo) 1101 { 1102 DEFINE_WAIT(wait); 1103 1104 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1105 for (;;) { 1106 if (!timeo) 1107 break; 1108 if (signal_pending(current)) 1109 break; 1110 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1111 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1112 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1113 break; 1114 if (sk->sk_shutdown & SEND_SHUTDOWN) 1115 break; 1116 if (sk->sk_err) 1117 break; 1118 timeo = schedule_timeout(timeo); 1119 } 1120 finish_wait(sk->sk_sleep, &wait); 1121 return timeo; 1122 } 1123 1124 1125 /* 1126 * Generic send/receive buffer handlers 1127 */ 1128 1129 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1130 unsigned long header_len, 1131 unsigned long data_len, 1132 int noblock, int *errcode) 1133 { 1134 struct sk_buff *skb; 1135 gfp_t gfp_mask; 1136 long timeo; 1137 int err; 1138 1139 gfp_mask = sk->sk_allocation; 1140 if (gfp_mask & __GFP_WAIT) 1141 gfp_mask |= __GFP_REPEAT; 1142 1143 timeo = sock_sndtimeo(sk, noblock); 1144 while (1) { 1145 err = sock_error(sk); 1146 if (err != 0) 1147 goto failure; 1148 1149 err = -EPIPE; 1150 if (sk->sk_shutdown & SEND_SHUTDOWN) 1151 goto failure; 1152 1153 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1154 skb = alloc_skb(header_len, gfp_mask); 1155 if (skb) { 1156 int npages; 1157 int i; 1158 1159 /* No pages, we're done... */ 1160 if (!data_len) 1161 break; 1162 1163 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1164 skb->truesize += data_len; 1165 skb_shinfo(skb)->nr_frags = npages; 1166 for (i = 0; i < npages; i++) { 1167 struct page *page; 1168 skb_frag_t *frag; 1169 1170 page = alloc_pages(sk->sk_allocation, 0); 1171 if (!page) { 1172 err = -ENOBUFS; 1173 skb_shinfo(skb)->nr_frags = i; 1174 kfree_skb(skb); 1175 goto failure; 1176 } 1177 1178 frag = &skb_shinfo(skb)->frags[i]; 1179 frag->page = page; 1180 frag->page_offset = 0; 1181 frag->size = (data_len >= PAGE_SIZE ? 1182 PAGE_SIZE : 1183 data_len); 1184 data_len -= PAGE_SIZE; 1185 } 1186 1187 /* Full success... */ 1188 break; 1189 } 1190 err = -ENOBUFS; 1191 goto failure; 1192 } 1193 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1194 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1195 err = -EAGAIN; 1196 if (!timeo) 1197 goto failure; 1198 if (signal_pending(current)) 1199 goto interrupted; 1200 timeo = sock_wait_for_wmem(sk, timeo); 1201 } 1202 1203 skb_set_owner_w(skb, sk); 1204 return skb; 1205 1206 interrupted: 1207 err = sock_intr_errno(timeo); 1208 failure: 1209 *errcode = err; 1210 return NULL; 1211 } 1212 1213 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1214 int noblock, int *errcode) 1215 { 1216 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1217 } 1218 1219 static void __lock_sock(struct sock *sk) 1220 { 1221 DEFINE_WAIT(wait); 1222 1223 for(;;) { 1224 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1225 TASK_UNINTERRUPTIBLE); 1226 spin_unlock_bh(&sk->sk_lock.slock); 1227 schedule(); 1228 spin_lock_bh(&sk->sk_lock.slock); 1229 if(!sock_owned_by_user(sk)) 1230 break; 1231 } 1232 finish_wait(&sk->sk_lock.wq, &wait); 1233 } 1234 1235 static void __release_sock(struct sock *sk) 1236 { 1237 struct sk_buff *skb = sk->sk_backlog.head; 1238 1239 do { 1240 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1241 bh_unlock_sock(sk); 1242 1243 do { 1244 struct sk_buff *next = skb->next; 1245 1246 skb->next = NULL; 1247 sk->sk_backlog_rcv(sk, skb); 1248 1249 /* 1250 * We are in process context here with softirqs 1251 * disabled, use cond_resched_softirq() to preempt. 1252 * This is safe to do because we've taken the backlog 1253 * queue private: 1254 */ 1255 cond_resched_softirq(); 1256 1257 skb = next; 1258 } while (skb != NULL); 1259 1260 bh_lock_sock(sk); 1261 } while((skb = sk->sk_backlog.head) != NULL); 1262 } 1263 1264 /** 1265 * sk_wait_data - wait for data to arrive at sk_receive_queue 1266 * @sk: sock to wait on 1267 * @timeo: for how long 1268 * 1269 * Now socket state including sk->sk_err is changed only under lock, 1270 * hence we may omit checks after joining wait queue. 1271 * We check receive queue before schedule() only as optimization; 1272 * it is very likely that release_sock() added new data. 1273 */ 1274 int sk_wait_data(struct sock *sk, long *timeo) 1275 { 1276 int rc; 1277 DEFINE_WAIT(wait); 1278 1279 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1280 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1281 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1282 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1283 finish_wait(sk->sk_sleep, &wait); 1284 return rc; 1285 } 1286 1287 EXPORT_SYMBOL(sk_wait_data); 1288 1289 /* 1290 * Set of default routines for initialising struct proto_ops when 1291 * the protocol does not support a particular function. In certain 1292 * cases where it makes no sense for a protocol to have a "do nothing" 1293 * function, some default processing is provided. 1294 */ 1295 1296 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1297 { 1298 return -EOPNOTSUPP; 1299 } 1300 1301 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1302 int len, int flags) 1303 { 1304 return -EOPNOTSUPP; 1305 } 1306 1307 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1308 { 1309 return -EOPNOTSUPP; 1310 } 1311 1312 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1313 { 1314 return -EOPNOTSUPP; 1315 } 1316 1317 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1318 int *len, int peer) 1319 { 1320 return -EOPNOTSUPP; 1321 } 1322 1323 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt) 1324 { 1325 return 0; 1326 } 1327 1328 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1329 { 1330 return -EOPNOTSUPP; 1331 } 1332 1333 int sock_no_listen(struct socket *sock, int backlog) 1334 { 1335 return -EOPNOTSUPP; 1336 } 1337 1338 int sock_no_shutdown(struct socket *sock, int how) 1339 { 1340 return -EOPNOTSUPP; 1341 } 1342 1343 int sock_no_setsockopt(struct socket *sock, int level, int optname, 1344 char __user *optval, int optlen) 1345 { 1346 return -EOPNOTSUPP; 1347 } 1348 1349 int sock_no_getsockopt(struct socket *sock, int level, int optname, 1350 char __user *optval, int __user *optlen) 1351 { 1352 return -EOPNOTSUPP; 1353 } 1354 1355 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1356 size_t len) 1357 { 1358 return -EOPNOTSUPP; 1359 } 1360 1361 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1362 size_t len, int flags) 1363 { 1364 return -EOPNOTSUPP; 1365 } 1366 1367 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 1368 { 1369 /* Mirror missing mmap method error code */ 1370 return -ENODEV; 1371 } 1372 1373 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 1374 { 1375 ssize_t res; 1376 struct msghdr msg = {.msg_flags = flags}; 1377 struct kvec iov; 1378 char *kaddr = kmap(page); 1379 iov.iov_base = kaddr + offset; 1380 iov.iov_len = size; 1381 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 1382 kunmap(page); 1383 return res; 1384 } 1385 1386 /* 1387 * Default Socket Callbacks 1388 */ 1389 1390 static void sock_def_wakeup(struct sock *sk) 1391 { 1392 read_lock(&sk->sk_callback_lock); 1393 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1394 wake_up_interruptible_all(sk->sk_sleep); 1395 read_unlock(&sk->sk_callback_lock); 1396 } 1397 1398 static void sock_def_error_report(struct sock *sk) 1399 { 1400 read_lock(&sk->sk_callback_lock); 1401 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1402 wake_up_interruptible(sk->sk_sleep); 1403 sk_wake_async(sk,0,POLL_ERR); 1404 read_unlock(&sk->sk_callback_lock); 1405 } 1406 1407 static void sock_def_readable(struct sock *sk, int len) 1408 { 1409 read_lock(&sk->sk_callback_lock); 1410 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1411 wake_up_interruptible(sk->sk_sleep); 1412 sk_wake_async(sk,1,POLL_IN); 1413 read_unlock(&sk->sk_callback_lock); 1414 } 1415 1416 static void sock_def_write_space(struct sock *sk) 1417 { 1418 read_lock(&sk->sk_callback_lock); 1419 1420 /* Do not wake up a writer until he can make "significant" 1421 * progress. --DaveM 1422 */ 1423 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 1424 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1425 wake_up_interruptible(sk->sk_sleep); 1426 1427 /* Should agree with poll, otherwise some programs break */ 1428 if (sock_writeable(sk)) 1429 sk_wake_async(sk, 2, POLL_OUT); 1430 } 1431 1432 read_unlock(&sk->sk_callback_lock); 1433 } 1434 1435 static void sock_def_destruct(struct sock *sk) 1436 { 1437 kfree(sk->sk_protinfo); 1438 } 1439 1440 void sk_send_sigurg(struct sock *sk) 1441 { 1442 if (sk->sk_socket && sk->sk_socket->file) 1443 if (send_sigurg(&sk->sk_socket->file->f_owner)) 1444 sk_wake_async(sk, 3, POLL_PRI); 1445 } 1446 1447 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1448 unsigned long expires) 1449 { 1450 if (!mod_timer(timer, expires)) 1451 sock_hold(sk); 1452 } 1453 1454 EXPORT_SYMBOL(sk_reset_timer); 1455 1456 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 1457 { 1458 if (timer_pending(timer) && del_timer(timer)) 1459 __sock_put(sk); 1460 } 1461 1462 EXPORT_SYMBOL(sk_stop_timer); 1463 1464 void sock_init_data(struct socket *sock, struct sock *sk) 1465 { 1466 skb_queue_head_init(&sk->sk_receive_queue); 1467 skb_queue_head_init(&sk->sk_write_queue); 1468 skb_queue_head_init(&sk->sk_error_queue); 1469 #ifdef CONFIG_NET_DMA 1470 skb_queue_head_init(&sk->sk_async_wait_queue); 1471 #endif 1472 1473 sk->sk_send_head = NULL; 1474 1475 init_timer(&sk->sk_timer); 1476 1477 sk->sk_allocation = GFP_KERNEL; 1478 sk->sk_rcvbuf = sysctl_rmem_default; 1479 sk->sk_sndbuf = sysctl_wmem_default; 1480 sk->sk_state = TCP_CLOSE; 1481 sk->sk_socket = sock; 1482 1483 sock_set_flag(sk, SOCK_ZAPPED); 1484 1485 if(sock) 1486 { 1487 sk->sk_type = sock->type; 1488 sk->sk_sleep = &sock->wait; 1489 sock->sk = sk; 1490 } else 1491 sk->sk_sleep = NULL; 1492 1493 rwlock_init(&sk->sk_dst_lock); 1494 rwlock_init(&sk->sk_callback_lock); 1495 lockdep_set_class(&sk->sk_callback_lock, 1496 af_callback_keys + sk->sk_family); 1497 1498 sk->sk_state_change = sock_def_wakeup; 1499 sk->sk_data_ready = sock_def_readable; 1500 sk->sk_write_space = sock_def_write_space; 1501 sk->sk_error_report = sock_def_error_report; 1502 sk->sk_destruct = sock_def_destruct; 1503 1504 sk->sk_sndmsg_page = NULL; 1505 sk->sk_sndmsg_off = 0; 1506 1507 sk->sk_peercred.pid = 0; 1508 sk->sk_peercred.uid = -1; 1509 sk->sk_peercred.gid = -1; 1510 sk->sk_write_pending = 0; 1511 sk->sk_rcvlowat = 1; 1512 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 1513 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 1514 1515 sk->sk_stamp.tv_sec = -1L; 1516 sk->sk_stamp.tv_usec = -1L; 1517 1518 atomic_set(&sk->sk_refcnt, 1); 1519 } 1520 1521 void fastcall lock_sock_nested(struct sock *sk, int subclass) 1522 { 1523 might_sleep(); 1524 spin_lock_bh(&sk->sk_lock.slock); 1525 if (sk->sk_lock.owner) 1526 __lock_sock(sk); 1527 sk->sk_lock.owner = (void *)1; 1528 spin_unlock(&sk->sk_lock.slock); 1529 /* 1530 * The sk_lock has mutex_lock() semantics here: 1531 */ 1532 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 1533 local_bh_enable(); 1534 } 1535 1536 EXPORT_SYMBOL(lock_sock_nested); 1537 1538 void fastcall release_sock(struct sock *sk) 1539 { 1540 /* 1541 * The sk_lock has mutex_unlock() semantics: 1542 */ 1543 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1544 1545 spin_lock_bh(&sk->sk_lock.slock); 1546 if (sk->sk_backlog.tail) 1547 __release_sock(sk); 1548 sk->sk_lock.owner = NULL; 1549 if (waitqueue_active(&sk->sk_lock.wq)) 1550 wake_up(&sk->sk_lock.wq); 1551 spin_unlock_bh(&sk->sk_lock.slock); 1552 } 1553 EXPORT_SYMBOL(release_sock); 1554 1555 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 1556 { 1557 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1558 sock_enable_timestamp(sk); 1559 if (sk->sk_stamp.tv_sec == -1) 1560 return -ENOENT; 1561 if (sk->sk_stamp.tv_sec == 0) 1562 do_gettimeofday(&sk->sk_stamp); 1563 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ? 1564 -EFAULT : 0; 1565 } 1566 EXPORT_SYMBOL(sock_get_timestamp); 1567 1568 void sock_enable_timestamp(struct sock *sk) 1569 { 1570 if (!sock_flag(sk, SOCK_TIMESTAMP)) { 1571 sock_set_flag(sk, SOCK_TIMESTAMP); 1572 net_enable_timestamp(); 1573 } 1574 } 1575 EXPORT_SYMBOL(sock_enable_timestamp); 1576 1577 /* 1578 * Get a socket option on an socket. 1579 * 1580 * FIX: POSIX 1003.1g is very ambiguous here. It states that 1581 * asynchronous errors should be reported by getsockopt. We assume 1582 * this means if you specify SO_ERROR (otherwise whats the point of it). 1583 */ 1584 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1585 char __user *optval, int __user *optlen) 1586 { 1587 struct sock *sk = sock->sk; 1588 1589 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1590 } 1591 1592 EXPORT_SYMBOL(sock_common_getsockopt); 1593 1594 #ifdef CONFIG_COMPAT 1595 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 1596 char __user *optval, int __user *optlen) 1597 { 1598 struct sock *sk = sock->sk; 1599 1600 if (sk->sk_prot->compat_getsockopt != NULL) 1601 return sk->sk_prot->compat_getsockopt(sk, level, optname, 1602 optval, optlen); 1603 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1604 } 1605 EXPORT_SYMBOL(compat_sock_common_getsockopt); 1606 #endif 1607 1608 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1609 struct msghdr *msg, size_t size, int flags) 1610 { 1611 struct sock *sk = sock->sk; 1612 int addr_len = 0; 1613 int err; 1614 1615 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 1616 flags & ~MSG_DONTWAIT, &addr_len); 1617 if (err >= 0) 1618 msg->msg_namelen = addr_len; 1619 return err; 1620 } 1621 1622 EXPORT_SYMBOL(sock_common_recvmsg); 1623 1624 /* 1625 * Set socket options on an inet socket. 1626 */ 1627 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1628 char __user *optval, int optlen) 1629 { 1630 struct sock *sk = sock->sk; 1631 1632 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1633 } 1634 1635 EXPORT_SYMBOL(sock_common_setsockopt); 1636 1637 #ifdef CONFIG_COMPAT 1638 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 1639 char __user *optval, int optlen) 1640 { 1641 struct sock *sk = sock->sk; 1642 1643 if (sk->sk_prot->compat_setsockopt != NULL) 1644 return sk->sk_prot->compat_setsockopt(sk, level, optname, 1645 optval, optlen); 1646 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1647 } 1648 EXPORT_SYMBOL(compat_sock_common_setsockopt); 1649 #endif 1650 1651 void sk_common_release(struct sock *sk) 1652 { 1653 if (sk->sk_prot->destroy) 1654 sk->sk_prot->destroy(sk); 1655 1656 /* 1657 * Observation: when sock_common_release is called, processes have 1658 * no access to socket. But net still has. 1659 * Step one, detach it from networking: 1660 * 1661 * A. Remove from hash tables. 1662 */ 1663 1664 sk->sk_prot->unhash(sk); 1665 1666 /* 1667 * In this point socket cannot receive new packets, but it is possible 1668 * that some packets are in flight because some CPU runs receiver and 1669 * did hash table lookup before we unhashed socket. They will achieve 1670 * receive queue and will be purged by socket destructor. 1671 * 1672 * Also we still have packets pending on receive queue and probably, 1673 * our own packets waiting in device queues. sock_destroy will drain 1674 * receive queue, but transmitted packets will delay socket destruction 1675 * until the last reference will be released. 1676 */ 1677 1678 sock_orphan(sk); 1679 1680 xfrm_sk_free_policy(sk); 1681 1682 sk_refcnt_debug_release(sk); 1683 sock_put(sk); 1684 } 1685 1686 EXPORT_SYMBOL(sk_common_release); 1687 1688 static DEFINE_RWLOCK(proto_list_lock); 1689 static LIST_HEAD(proto_list); 1690 1691 int proto_register(struct proto *prot, int alloc_slab) 1692 { 1693 char *request_sock_slab_name = NULL; 1694 char *timewait_sock_slab_name; 1695 int rc = -ENOBUFS; 1696 1697 if (alloc_slab) { 1698 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 1699 SLAB_HWCACHE_ALIGN, NULL, NULL); 1700 1701 if (prot->slab == NULL) { 1702 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n", 1703 prot->name); 1704 goto out; 1705 } 1706 1707 if (prot->rsk_prot != NULL) { 1708 static const char mask[] = "request_sock_%s"; 1709 1710 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1711 if (request_sock_slab_name == NULL) 1712 goto out_free_sock_slab; 1713 1714 sprintf(request_sock_slab_name, mask, prot->name); 1715 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name, 1716 prot->rsk_prot->obj_size, 0, 1717 SLAB_HWCACHE_ALIGN, NULL, NULL); 1718 1719 if (prot->rsk_prot->slab == NULL) { 1720 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n", 1721 prot->name); 1722 goto out_free_request_sock_slab_name; 1723 } 1724 } 1725 1726 if (prot->twsk_prot != NULL) { 1727 static const char mask[] = "tw_sock_%s"; 1728 1729 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1730 1731 if (timewait_sock_slab_name == NULL) 1732 goto out_free_request_sock_slab; 1733 1734 sprintf(timewait_sock_slab_name, mask, prot->name); 1735 prot->twsk_prot->twsk_slab = 1736 kmem_cache_create(timewait_sock_slab_name, 1737 prot->twsk_prot->twsk_obj_size, 1738 0, SLAB_HWCACHE_ALIGN, 1739 NULL, NULL); 1740 if (prot->twsk_prot->twsk_slab == NULL) 1741 goto out_free_timewait_sock_slab_name; 1742 } 1743 } 1744 1745 write_lock(&proto_list_lock); 1746 list_add(&prot->node, &proto_list); 1747 write_unlock(&proto_list_lock); 1748 rc = 0; 1749 out: 1750 return rc; 1751 out_free_timewait_sock_slab_name: 1752 kfree(timewait_sock_slab_name); 1753 out_free_request_sock_slab: 1754 if (prot->rsk_prot && prot->rsk_prot->slab) { 1755 kmem_cache_destroy(prot->rsk_prot->slab); 1756 prot->rsk_prot->slab = NULL; 1757 } 1758 out_free_request_sock_slab_name: 1759 kfree(request_sock_slab_name); 1760 out_free_sock_slab: 1761 kmem_cache_destroy(prot->slab); 1762 prot->slab = NULL; 1763 goto out; 1764 } 1765 1766 EXPORT_SYMBOL(proto_register); 1767 1768 void proto_unregister(struct proto *prot) 1769 { 1770 write_lock(&proto_list_lock); 1771 list_del(&prot->node); 1772 write_unlock(&proto_list_lock); 1773 1774 if (prot->slab != NULL) { 1775 kmem_cache_destroy(prot->slab); 1776 prot->slab = NULL; 1777 } 1778 1779 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 1780 const char *name = kmem_cache_name(prot->rsk_prot->slab); 1781 1782 kmem_cache_destroy(prot->rsk_prot->slab); 1783 kfree(name); 1784 prot->rsk_prot->slab = NULL; 1785 } 1786 1787 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 1788 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab); 1789 1790 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 1791 kfree(name); 1792 prot->twsk_prot->twsk_slab = NULL; 1793 } 1794 } 1795 1796 EXPORT_SYMBOL(proto_unregister); 1797 1798 #ifdef CONFIG_PROC_FS 1799 static inline struct proto *__proto_head(void) 1800 { 1801 return list_entry(proto_list.next, struct proto, node); 1802 } 1803 1804 static inline struct proto *proto_head(void) 1805 { 1806 return list_empty(&proto_list) ? NULL : __proto_head(); 1807 } 1808 1809 static inline struct proto *proto_next(struct proto *proto) 1810 { 1811 return proto->node.next == &proto_list ? NULL : 1812 list_entry(proto->node.next, struct proto, node); 1813 } 1814 1815 static inline struct proto *proto_get_idx(loff_t pos) 1816 { 1817 struct proto *proto; 1818 loff_t i = 0; 1819 1820 list_for_each_entry(proto, &proto_list, node) 1821 if (i++ == pos) 1822 goto out; 1823 1824 proto = NULL; 1825 out: 1826 return proto; 1827 } 1828 1829 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 1830 { 1831 read_lock(&proto_list_lock); 1832 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN; 1833 } 1834 1835 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1836 { 1837 ++*pos; 1838 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v); 1839 } 1840 1841 static void proto_seq_stop(struct seq_file *seq, void *v) 1842 { 1843 read_unlock(&proto_list_lock); 1844 } 1845 1846 static char proto_method_implemented(const void *method) 1847 { 1848 return method == NULL ? 'n' : 'y'; 1849 } 1850 1851 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 1852 { 1853 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s " 1854 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 1855 proto->name, 1856 proto->obj_size, 1857 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1, 1858 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1, 1859 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI", 1860 proto->max_header, 1861 proto->slab == NULL ? "no" : "yes", 1862 module_name(proto->owner), 1863 proto_method_implemented(proto->close), 1864 proto_method_implemented(proto->connect), 1865 proto_method_implemented(proto->disconnect), 1866 proto_method_implemented(proto->accept), 1867 proto_method_implemented(proto->ioctl), 1868 proto_method_implemented(proto->init), 1869 proto_method_implemented(proto->destroy), 1870 proto_method_implemented(proto->shutdown), 1871 proto_method_implemented(proto->setsockopt), 1872 proto_method_implemented(proto->getsockopt), 1873 proto_method_implemented(proto->sendmsg), 1874 proto_method_implemented(proto->recvmsg), 1875 proto_method_implemented(proto->sendpage), 1876 proto_method_implemented(proto->bind), 1877 proto_method_implemented(proto->backlog_rcv), 1878 proto_method_implemented(proto->hash), 1879 proto_method_implemented(proto->unhash), 1880 proto_method_implemented(proto->get_port), 1881 proto_method_implemented(proto->enter_memory_pressure)); 1882 } 1883 1884 static int proto_seq_show(struct seq_file *seq, void *v) 1885 { 1886 if (v == SEQ_START_TOKEN) 1887 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 1888 "protocol", 1889 "size", 1890 "sockets", 1891 "memory", 1892 "press", 1893 "maxhdr", 1894 "slab", 1895 "module", 1896 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 1897 else 1898 proto_seq_printf(seq, v); 1899 return 0; 1900 } 1901 1902 static struct seq_operations proto_seq_ops = { 1903 .start = proto_seq_start, 1904 .next = proto_seq_next, 1905 .stop = proto_seq_stop, 1906 .show = proto_seq_show, 1907 }; 1908 1909 static int proto_seq_open(struct inode *inode, struct file *file) 1910 { 1911 return seq_open(file, &proto_seq_ops); 1912 } 1913 1914 static const struct file_operations proto_seq_fops = { 1915 .owner = THIS_MODULE, 1916 .open = proto_seq_open, 1917 .read = seq_read, 1918 .llseek = seq_lseek, 1919 .release = seq_release, 1920 }; 1921 1922 static int __init proto_init(void) 1923 { 1924 /* register /proc/net/protocols */ 1925 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0; 1926 } 1927 1928 subsys_initcall(proto_init); 1929 1930 #endif /* PROC_FS */ 1931 1932 EXPORT_SYMBOL(sk_alloc); 1933 EXPORT_SYMBOL(sk_free); 1934 EXPORT_SYMBOL(sk_send_sigurg); 1935 EXPORT_SYMBOL(sock_alloc_send_skb); 1936 EXPORT_SYMBOL(sock_init_data); 1937 EXPORT_SYMBOL(sock_kfree_s); 1938 EXPORT_SYMBOL(sock_kmalloc); 1939 EXPORT_SYMBOL(sock_no_accept); 1940 EXPORT_SYMBOL(sock_no_bind); 1941 EXPORT_SYMBOL(sock_no_connect); 1942 EXPORT_SYMBOL(sock_no_getname); 1943 EXPORT_SYMBOL(sock_no_getsockopt); 1944 EXPORT_SYMBOL(sock_no_ioctl); 1945 EXPORT_SYMBOL(sock_no_listen); 1946 EXPORT_SYMBOL(sock_no_mmap); 1947 EXPORT_SYMBOL(sock_no_poll); 1948 EXPORT_SYMBOL(sock_no_recvmsg); 1949 EXPORT_SYMBOL(sock_no_sendmsg); 1950 EXPORT_SYMBOL(sock_no_sendpage); 1951 EXPORT_SYMBOL(sock_no_setsockopt); 1952 EXPORT_SYMBOL(sock_no_shutdown); 1953 EXPORT_SYMBOL(sock_no_socketpair); 1954 EXPORT_SYMBOL(sock_rfree); 1955 EXPORT_SYMBOL(sock_setsockopt); 1956 EXPORT_SYMBOL(sock_wfree); 1957 EXPORT_SYMBOL(sock_wmalloc); 1958 EXPORT_SYMBOL(sock_i_uid); 1959 EXPORT_SYMBOL(sock_i_ino); 1960 EXPORT_SYMBOL(sysctl_optmem_max); 1961 #ifdef CONFIG_SYSCTL 1962 EXPORT_SYMBOL(sysctl_rmem_max); 1963 EXPORT_SYMBOL(sysctl_wmem_max); 1964 #endif 1965